Muse Glimmer 30B ROCmFP4 Strix Halo DFlash needs about 15.7 GB of memory at its main file (a 15.0 GB file) with an 8K context window, so it fits on a 16 GB GPU. This repository has 8 quants, from 1.5 GB to 29.2 GB (Q8_0). Longer context needs more: at 32K tokens the KV cache alone adds about 0.7 GB.
Muse Glimmer 30B ROCmFP4 Strix Halo DFlash – a multimodal, agentic LLM for local use
Muse Glimmer 30B is a 28‑billion‑parameter causal language model with a dedicated perception encoder. It supports reasoning, tool use, vision, and multi‑step workflows. The model is optimized for local deployment on AMD hardware with ROCmFP4 tensors.
Good at
- Multimodal reasoning with text and images
- Reliable tool invocation and error recovery
- Long‑context agentic task completion
Limits
- Requires ROCmFPX fork of llama.cpp
- No official support for other architectures
- Large VRAM needed for full precision
Best for
- Building autonomous agents on consumer GPUs
- Local multimodal chat applications
- Debugging and code generation workflows
Who it’s for. Developers and researchers who need a fully local, multimodal LLM that can run on AMD Ryzen AI Max+ 395 or Radeon 8060S GPUs. The model is suitable for building agents that perform complex tasks without cloud access.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 1.5 GB | 2.2 GB | Fits | Get file → | |
| — | 1.6 GB | 2.3 GB | Fits | Get file → | |
| — | 14.8 GB | 15.5 GB | Too big | Get file → | |
| — | 15.0 GB | 15.7 GB | Too big | Get file → | |
| — | 15.2 GB | 15.9 GB | Too big | Get file → | |
| — | 18.1 GB | 18.8 GB | Too big | Get file → | |
| Q8_0 | 28.8 GB | 29.5 GB | Too big | Get file → | |
| Q8_0 | 29.2 GB | 29.9 GB | Too big | Get file → |
For 8 GB VRAM, use the 1.49 GB or 1.63 GB unknown‑quant files; memory usage is approximately 2.2 GB or 2.3 GB at 8 k context. For 16 GB VRAM, the 28.83 GB Q8_0 file is recommended; it uses about 29.5 GB of memory at 8 k context. The 29.24 GB Q8_0 file uses roughly 29.9 GB of memory.
What GPU do you need for Muse Glimmer 30B ROCmFP4 Strix Halo DFlash?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: the best quant that fits is the main file (1.6 GB file, about 2.3 GB in use).
- 6 GB VRAM: the best quant that fits is the main file (1.6 GB file, about 2.3 GB in use).
- 8 GB VRAM: the best quant that fits is the main file (1.6 GB file, about 2.3 GB in use).
- 12 GB VRAM: the best quant that fits is the main file (1.6 GB file, about 2.3 GB in use).
- 16 GB VRAM: the best quant that fits is the main file (15.2 GB file, about 15.9 GB in use).
- 24 GB VRAM: the best quant that fits is the main file (18.1 GB file, about 18.8 GB in use).
- 32 GB VRAM: the best quant that fits is Q8_0 (29.2 GB file, about 29.9 GB in use).
Which Muse Glimmer 30B ROCmFP4 Strix Halo DFlash quant should you choose?
If memory is tight, the smallest file is the first one in the table at 1.5 GB, with the quality trade-offs that come with fewer bits.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Muse Glimmer 30B ROCmFP4 Strix Halo DFlash
How much VRAM does Muse Glimmer 30B ROCmFP4 Strix Halo DFlash need?
About 15.7 GB at its main file with an 8K context window. That is the 15.0 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run Muse Glimmer 30B ROCmFP4 Strix Halo DFlash on an 8 GB GPU?
Yes, with the smaller files: the 1.6 GB file needs about 2.3 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Muse Glimmer 30B ROCmFP4 Strix Halo DFlash GGUF?
It depends on the quant: file is 1.5 GB, file is 1.6 GB, file is 14.8 GB, file is 15.0 GB, file is 15.2 GB, file is 18.1 GB and 2 more.
What context length does Muse Glimmer 30B ROCmFP4 Strix Halo DFlash support?
The GGUF header lists a trained context length of 131,072 tokens (128K). The KV cache grows with the context you use: about 0.2 GB at 8K and 0.7 GB at 32K.
Where can I download Muse Glimmer 30B ROCmFP4 Strix Halo DFlash GGUF?
The files are on Hugging Face in the kingjones777/Muse-Glimmer-30B-ROCmFP4-Strix-Halo-DFlash-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does Muse Glimmer 30B ROCmFP4 Strix Halo DFlash use?
Hugging Face lists the license as other. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run Muse Glimmer on 8 GB VRAM?
Yes, using the 1.49 GB or 1.63 GB unknown‑quant files. Estimated VRAM usage is 2.2 GB or 2.3 GB at 8 k context.
Does this model support vision?
Yes, it includes a dedicated perception encoder for interleaved text and image input.
What runtime is required?
The model needs the ROCmFPX fork of llama.cpp with the muse‑glimmer architecture and ROCmFP4 tensor types.